Use normal sampling for SAC

docs/misc/changelog.rst

@@ -21,6 +21,7 @@ Deprecations:
 
 Others:
 ^^^^^^^
+- SAC with SDE now sample only one matrix
 
 Documentation:
 ^^^^^^^^^^^^^^

setup.py

@@ -47,7 +47,7 @@ setup(name='torchy_baselines',
       license="MIT",
       long_description="",
       long_description_content_type='text/markdown',
-      version="0.2.0",
+      version="0.2.1",
       )
 
 # python setup.py sdist

torchy_baselines/__init__.py

@@ -4,4 +4,4 @@ from torchy_baselines.ppo import PPO
 from torchy_baselines.sac import SAC
 from torchy_baselines.td3 import TD3
 
-__version__ = "0.2.0"
+__version__ = "0.2.1"

torchy_baselines/sac/sac.py

@@ -171,8 +171,8 @@ class SAC(OffPolicyRLModel):
             # is lost and we cannot backpropagate through again
             # anyway, we need to sample because `log_std` may have changed between two gradient steps
             if self.use_sde:
-                self.actor.reset_noise(batch_size=batch_size)
-                # self.actor.reset_noise()
+                # self.actor.reset_noise(batch_size=batch_size)
+                self.actor.reset_noise()
 
             # Action by the current actor for the sampled state
             action_pi, log_prob = self.actor.action_log_prob(obs)